Phantom Quadcopter

It appears as a UFO to the uninitiated and it is only somewhat of this world—it is a Phantom. The Phantom quadcopter produced by DJI Innovations is the first high-quality, medium-cost consumer quadcopter to come to market. There have been other high-end multirotor aircraft available for the past 5 or so years, but with costs measured in the thousands to tens of thousands of dollars, they were aimed squarely at the film production industry rather than at hobbyists.

I first got into the multirotor game a year and a half ago when I built an ArduCopter-brand quadcopter from 3DRobotics in hopes of creating dramatic video footage in Antarctica. But despite repeated attempts and diligence on my part, the ArduCopter has been nothing more than a source of frustration and work. The software that controls the craft is open-source, which means that people from all over the world can contribute to it in the same way that open-source has been used to build the Unix operating system and thousands of other web tools. But there is a major difference between a web browser plug-in and a helicopter; if your web browser crashes due to a poorly tested plug-in, you simply disable the plug-in and restart the browser. But when your helicopter crashes, you are literally picking up the pieces and are instantly hundreds of dollars poorer than seconds before. This is precisely what I experienced with the ArduCopter again and again.

Quadcopters (one type of multicopter which includes tricopters, Y-copters, hexacopters, octocopters, etc.) maintain control based on an entirely different principle than traditional helicopters. Normal helicopters use a very complex set of mechanical levers  to precisely control the tilt of the main rotor blades in order to tilt the aircraft forward or back (pitch) and left or right (roll) and also increase or decrease the total lift. The last degree of freedom (yaw/heading) is achieved by increasing or decreasing the horizontal thrust on the tail rotor. Since their inception, helicopters have been notoriously difficult to fly whether they were full-size or small remote control (RC) models. For this reason, it was a rare sight to see an RC helicopter in your daily life and their costs were beyond all but the most adamant hobbyists.

In comparison to the complex mechanical systems of the traditional helicopter, multirotors use a much simpler method to achieve control—differential thrust. In the case of a quadcopter, the craft will pitch forward when the rear rotors are producing more thrust (spinning faster) than the front rotors, and similarly, it will roll to the right when the left rotors are producing more thrust than the right rotors. Yaw control is a little bit more subtle, the adjacent rotors spin in opposite directions so the net torque on the craft is nulled, and heading changes are created by increasing the speed of either the clockwise or counterclockwise rotors in relation to their counterparts.

In principle, this all sounds very simple and you might wonder why multirotors are just beginning to take off both literally and figuratively. For me, this is where the story gets really interesting. During my time at MIT in the late ‘90s, I worked on my Master’s thesis on a project for the Defense Advanced Research Projects Agency (DARPA) on a concept called Micro Air Vehicles (MAVs). The goal was to build miniature, autonomous aircraft that soldiers could deploy from their backpacks for reconnaissance . I was working at Draper Laboratories who is a leader in guidance and control systems and one of the initial developers of micro-electromechanical systems (MEMS) gyroscopes and accelerometers. Draper was selected by DARPA for the MAV project because one of the key barriers to overcome was building very small electronics to stabilize the craft, and the MEMS gyros and accelerometers in combination with decades of designing flight controllers were just what the doctor ordered. My Master’s thesis research was spent developing robust control algorithms for a single rotor MAV that unfortunately stopped receiving funding a year after I began full time employment at Draper. Without the MAV work at Draper, my professional interests in Cambridge waned and I headed west in 2001 to Boulder and forgot about the MAV for many years.

But in the early 2000’s, something very interesting began to take place in the consumer electronics market that would ultimately result in the feasibility of the Phantom. Cell phones began to fill every pocket and purse, and GPS units started populating the dashboards of our cars. And eventually, cell phones became smart phones with GPS and MEMS gyros and accelerometers. All of a sudden, the key electronic components for a flight stabilization system were extremely cheap, and both hobbyists and entrepreneurs began to build stabilizing autopilots for remote control aircraft out of these low cost components. RC planes were initially the craft of choice due to their inherently stable flight characteristics, but creativity quickly followed and the multirotor phenomenon was born.

The reason why these sensors and microcontrollers were so important to the development of multirotors has to do with the concept of controllability. If you could imagine the simplest way to control a quadcopter, you would simply increase or decrease the speed of each of the rotors individually in order to accomplish your goal. For example, if you wanted to move the copter to the right, you would need to roll it to the right, and so you would need to increase the speed of the rotors on the left. So if you had four channels on your RC transmitter, you would need to increase the output on two of the channels but not the other two. And just as importantly, you would need to quickly decrease those values or it would do a barrel roll and end up crashing to the ground. Although it would be possible for a pilot to control the copter in this manner, it would be extremely challenging to manage all four channels simultaneously since the inputs (rotor speeds) and the outputs (copter orientation, speed, and position) are not directly related in a one-to-one fashion. What is missing in this manual scenario are the mixing and stabilizing functions that allow the inputs and outputs to be related in a one-to-one manner. And in order to implement these functions, a microcontroller (for performing all of the math and logic) and sensors (for measuring the orientation angles and position of the craft) are needed and are exactly what became affordable due to smart phone proliferation around 2010.

At about that time, the loose nit online community at DIYDrones.com started to apply their ArduPilot system to multirotors and shortly thereafter, the ArduCopter was released by the associated 3DRobotics store. I purchased one of these ArudCopters in October 2011 and dreamed it would manifest the control algorithms that I had designed for the DARPA MAV a decade earlier. As a hobby project, there was a reasonable amount of effort involved to assemble and configure the craft before I could fly it. The circuit boards, speed controllers, and motors all required soldering, the entire assembly needed to be screwed together, and software needed to be loaded onto the controller. But the biggest time sink and arguably the most important step was to properly configure the copter for flight. That meant that all of the electrical connections needed to perfectly coincide with how the software and remote control expected them to be connected. Documentation was provided which was typically vague and often flat out wrong, but I methodically proceeded to verify that everything was configured properly and finally was able to move forward with some test flights.

My very first flights were successful but a bit shaky. I suspected that the copter needed some tuning, but without fully understanding the underlying control algorithms, adjustments were a purely ad hoc experience and didn’t net any significant improvements. Then things got interesting. After initially flying in “Stabilize” mode, I began to test the more advanced flight modes that included increasing levels of autonomy. “Altitude Hold” mode was the next most logical step, but each time I enabled the mode, I lost complete control and the copter would veer out of control and eventually crash into the ground. I also briefly tested the “Position Hold” mode that utilized GPS, but it also behaved erratically with no success. All in all, I had a 50% success ratio with the ArduCopter—one flight would go pretty well, and the next flight would result in a horrible crash and replacement of propellers and motors. I checked that all of the sensors were performing correctly and I knew that the motors worked correctly based on good performance in “Stabilize” mode, so that just left the control algorithms as a suspect. Initially, I thought that my problems presented a perfect opportunity to apply my expertise in control systems to help make the ArduCopter project better, but after I dug around in the code and got to know the project better, I came to realize that the ArduCopter was well beyond the effort that I was willing to commit. The open-source nature of the project meant that users were contributing new code and algorithms from all around the world without any formal vetting or verification. Entirely new control topologies were being introduced without any formal analysis or simulation that proved their worth, and it was up to the users to determine whether the new approaches were valid or not via their flights tests and crashes. Although it is a very interesting approach, it is in direct contradiction to the conservative design-analysis-simulation-test approach that has been successfully developed in the aerospace industry over the past half century. So although I desperately wanted the ArduCopter project to work for me, I ultimately concluded that the underlying development approach was inconsistent with my goals of wanting a quadcopter that just worked—period.

Then a few months ago, I started coming across articles from varied web sources about a fantastic new quadcopter that was designed to specifically carry a GoPro for low cost cinematography. After reading dozens of these articles with authors extoling its virtues that included almost zero assembly, perfect flight right out of the box, simple operation for brand new pilots, and polished GPS integration, I began to pay serious attention to this new kid on the block. I started to remember my dreams of achieving dramatic aerial video footage on my adventures, and I thought that this new quadcopter might be just what I envisioned.

On May 7, I received the package in the mail—a shiny white DJI Phantom that looked like it was designed by Apple. Assembly was just as easy as reported—about 10 minutes. Configuration was trivial—I moved the copter in a circular motion to calibrate the compass. And the initial flight? Well, the initial test flight was exactly as stated—extremely stable with GPS-based position and altitude hold that worked right out of the box! On the second flight, I attached my GoPro, flew a little more aggressively, and even demonstrated how well “Position Hold” worked as I recorded video footage of the Phantom with my hands completely off the control sticks! I compiled the footage and included a neat picture-in-picture of the hands-off flight which you can see below:

Since those initial flights, I have been incrementally pushing the limits of the Phantom to learn its capabilities and how its videos will fit into my stories of adventure. Compared to my 50% success ratio with the ArduCopter, I have achieved 100% success over the first 10 flights of the Phantom. I attribute the successful performance of the Phantom to a simple yet critical characteristic that the ArduCopter lacked—complete and tight integration of all design and test aspects by a single organization which just also happens to be a core tenet of none other than Apple. What has been interesting about this endeavor is that although I expected it to teach me a lot about flight controls, more than anything else, it has been a philosophical lesson about systems integration.

In the year to come, expect lots of new stories and videos that involve footage from the Phantom, because this thing is grawesome!

Update, 9/4/2013

I have had tremendous success with the Phantom over the past four months and have used it to create several videos and capture some good photos. I am so enthralled with it that I invested in an FPV setup and a stabilizing camera gimbal to help get even better imagery.

And over Labor Day weekend, I finally got around to testing the Return to Home (RTH) functionality. After flying the copter 100m away, I turned off the transmitter and rather than immediately crashing to the ground, the RTH safety mode automatically kicked in flying it back to me and landing exactly where it took off. I am incredibly impressed with this little copter. Check out the video of the RTH in action: